Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
net/publication/315708427
CITATIONS READS
0 292
5 authors, including:
Bahar Dadashova
Texas A&M University
16 PUBLICATIONS 18 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
FURGOSEG Joint Project. P 24/08. Ministry of Public Works in its National Plan R&D 2008-2011 View project
Innovative Tools and Techniques in Identifying Highway Safety Improvement Projects: Technical Report View project
All content following this page was uploaded by Bahar Dadashova on 30 January 2019.
Abstract
Texas frontage road turnarounds at diamond interchange locations are a common treatment in the State of Texas. This
configuration, also often referred to as the Texas U-turn, allows vehicles traveling on a one-way frontage road to access
the opposing direction one-way frontage road via a U-turn before the terminal intersections at the diamond interchange.
This technique removes two potential left-turn maneuvers from the adjacent at-grade intersections. The frontage road
turnaround has operational benefits that result from shifting vehicles that would typically be occupying the intersection.
The safety of these turnaround configurations, however, is not well known. This paper focuses on the observed safety
performance of these turnaround configurations at diamond freeway interchanges in Texas. The authors first developed a
large randomly sampled data set to facilitate a statistically reliable assessment of U-turn safety performance for Texas
interchanges. Next they conducted a safety performance assessment at locations with and without turnarounds and
determined that site features that significantly influence the number of crashes include the volume of cross street traffic,
the cross street right-turn configuration at frontage roads, the minimum radius in the turnaround, the longitudinal dis-
tance from the U-turn exit to the closest downstream driveway, and the number of lanes on each frontage road. The
safety performance findings at locations with and without turnarounds suggests that an agency could construct a turn-
around to enhance facility operations with the knowledge that construction will not adversely impact safety. Although
the research introduced in this paper focuses on Texas locations, the findings should be applicable to similar facilities in
other states if the turnaround conditions are similar.
Frontage road turnarounds are located at major safety research consistent with the techniques recom-
interchanges in Texas in urban and suburban regions (1). mended in the Highway Safety Manual is not yet available
For a turnaround to be a feasible option, the frontage for the Texas turnaround configuration. Consequently,
roads located parallel to the freeway lanes must have transportation professionals do not have the ability to
one-way operations. At locations where the freeway directly assess the number of expected crashes at an inter-
lanes are depressed, the turnaround is located on a change location or determine how the configuration of the
structure adjacent to the arterial cross street lanes. When U-turn can influence the number of crashes. The research
the freeway is elevated, the turnaround location will be summarized in this paper, therefore, presents the results of
under a bridge as shown in Figure 1. For both a recent Texas turnaround safety assessment performed on
conditions, sight distance and visibility are important behalf of the Texas Department of Transportation
considerations that influence the U-turn geometry. (TxDOT). Although the analysis focuses on Texas inter-
In 2010, the American Association of State Highway changes, the findings from this study are likely to be
and Transportation Officials (AASHTO) published the
first edition of the Highway Safety Manual (2). This docu-
ment introduced the concept of integrating safety predic-
1
tion based on statistically reliable safety performance Texas A&M Transportation Institute, College Station, TX
functions that uniquely assess key characteristics of a road-
Corresponding Author:
way countermeasure or configuration. Unfortunately, Address correspondence to Karen K. Dixon: k-dixon@tti.tamu.edu
2 Transportation Research Record 00(0)
representative of similar interchange configurations in intersections for the selected roadway network (freeway
other states. and arterial).
Following this initial GIS screening for intersecting
points, the authors next matched the large number of
Database Development intersecting points (11,289 identified points) to potential
locations where an interchange was likely. To narrow the
Currently, the State of Texas does not maintain a com-
search, an intersection with multiple points (generally
prehensive intersection or interchange database. To over-
four to six points) located close together often repre-
come this limitation, the authors first developed a study
sented some sort of interchange. As can be observed in
sample by identifying candidate study sites using geo-
Figure 3a, an interchange with a U-turn will normally
graphic information system (GIS) merging techniques
have at least six intersecting points. Therefore, to identify
and then developing a geographically stratified random
these potential interchange locations, the authors applied
sample for database development and subsequent safety
an additional 300 ft (approximately 100 m) buffer to all
assessment.
intersecting points as shown in Figure 3b to measure the
distance between two intersecting points of randomly
selected interchanges. As the objective of this activity
Identifying Candidate Study Sites
was to include interchanges with and without U-turns, a
The authors used existing roadway functional system diamond interchange with four or more intersection
information identified in the TxDOT roadway inven- points within this buffer region represented a potential
tory database as an initial step toward determining free- U-turn location. Finally, the authors filtered locations
way and arterial networks where diamond interchanges and assigned a single consolidated intersecting point to
could be potentially located. Using GIS tools, the represent the latitude and longitude for the potential
authors identified locations where interstate, freeway, study interchange.
or expressway facilities crossed arterial roadways. The
authors developed separate GIS shape files based on
these road categories and identified locations for poten- Developing a Stratified Random Sample
tial points of the intersection that could represent inter- The selection of interchange locations that represent the
changes or intersections. These potential intersections larger diamond interchange population requires a ran-
included intersecting points as far as 200 ft (approxi- dom selection of the study locations. As the State of
mately 60 m) from the freeway centerline. Using manual Texas is very large, the sampling procedure considered
evaluations, the authors then confirmed which inter- geographic location so that the resulting sample could be
secting points did represent intersections between free- representative of the diamond interchanges across the
ways and arterials. Figure 2 depicts these identified state. The final TxDOT project report provides
Dixon et al 3
Figure 2. Locating an interchange with a U-turn (a) aerial view and (b) applying 300 ft buffer.
considerable detail about this sampling technique, but that included geometric data acquired from the TxDOT
the final stratified random sample included 168 diamond roadway inventory database and was supplemented or
interchanges with U-turns and 60 prospective diamond confirmed using aerials from Google Earth ProÒ. In
interchanges without U-turns. The authors used this data addition, the database included average annual daily
set for the subsequent site selection and data collection traffic (AADT), the K-factor (for converting daily traffic
activities. The geographically stratified random sample volume proportions during peak hours), and the direc-
included a two-stage process in which the authors first tional distribution factor known as the D-factor. This D-
randomly selected interchanges from the entire state and factor proved to be a valuable way to confirm one-way
then randomly selected additional interchanges so that frontage road operations. In addition, the authors
the dataset included at least one interchange from every assigned crash data acquired from the Texas Crash
TxDOT district with freeway facilities. The initial inter- Records Information System for the years 2009–2015.
change identification utilized geographic shapefiles and The randomly selected study sites included a mixture
locations where arterials and freeway vectors crossed fol- of locations with and without U-turns. For locations
lowed by photographic confirmation of each diamond with U-turns, the authors reviewed historic aerial photo-
interchange. graphs to confirm the presence of the U-turn for the
entire seven-year period. For locations where the U-turn
construction occurred during this study period, the
Data Collection and Database Development authors removed any unconfirmed crash years. As an
Following the identification of the prospective study example, if an aerial photograph indicated a U-turn was
sites, the authors acquired data for each candidate site not present in 2010 but was present in 2013, the database
4 Transportation Research Record 00(0)
Figure 3. Locating an interchange with a U-turn (a) aerial view and (b) applying 300 ft buffer.
excluded crash data for 2011 and 2012. In addition, forward as a half site), the authors determined the num-
depending on the date of the aerial photograph, 2010 ber of frontage road main lanes at the intersection
and 2013 crash data may have also been removed if the approach. Figure 4 depicts the total number of lanes as
‘‘before’’ site aerial was not in December or the ‘‘after’’ FRA and FRB. The authors used the labels ‘‘A’’ and ‘‘B’’
site aerial was not in January of the study year. to identify each half site for documentation purposes.
The analysis database included many potential site They do not imply specific directional information. It
variables. Many of these did not prove to be significant should be noted that most frontage roads had two to
and are not included for the purposes of brevity (includ- three lanes near the U-turn locations. It should also be
ing frontage road AADT, which the research team deter- noted that the study included 203 half sites with turn-
mined provided inaccurate values when compared to the arounds and 133 half sites that did not have turnarounds.
crossroad AADT). Key data elements influential to the This resulted in a total of 336 half sites.
safety performance include the following data elements:
Number of total lanes on each frontage road Longitudinal Distance Between the U-turn Exit and the First
approach; Downstream Driveway. The longitudinal distance to the first
Longitudinal distance between the U-turn exit and downstream driveway, measured from the point on the
the first downstream driveway; frontage road where the U-turning vehicles exit the U-turn
Presence of traffic signal at the study intersections. and merge onto the frontage road, can influence the weav-
Arterial right-turn treatment onto frontage roads. ing and merging conditions on the frontage road. This
Depressed or elevated U-turn configuration; and label is represented as DWYA and DWYB in Figure 4. For
U-turn leg dimensions (radii, lane widths, and this study, the authors measured this distance for the drive-
length of dedicated U-turn). ways that were within 500 ft of the U-turn exit. Out of 203
half sites, 53 of the locations had driveways positioned out-
In some cases, the data reduction process could not side of this 500 ft threshold. The authors assigned a default
determine all site characteristics for a location. Often this DWY value of 500 ft for these locations.
constraint was a result of the interchange configuration.
For example, U-turn leg dimensions could not always be
Presence of a Traffic Signal at Study Intersection. The authors
determined at underpass turnaround locations because of
used aerial photographs and the companion StreetView
occlusion from the structure. The authors elected to retain
tool to determine the type of traffic control present at
these sites and use a varying sample size during the statis-
each frontage road and cross street intersection location.
tical analysis stepwise variable assessment. The following
At locations with a traffic signal, this method did not per-
sections review each of these roadway characteristics.
mit acquiring signal timing information. Approximately
93% of the intersections had traffic signals. Ultimately,
Number of Total Lanes on Each Frontage Road Approach. For the authors included only signalized intersection loca-
each side of a study site (referred to from this point tions in the subsequent safety assessment.
Dixon et al 5
Arterial Right-Turn Treatment onto Frontage Roads. As a vehi- arterial crossroads located below a freeway bridge, and
cle that is turning right from the cross street onto the one- 36 (18 of 168 sites) were elevated above the freeway.
way frontage road may potentially encounter a conflict Although the placement as elevated versus depressed did
with a vehicle exiting the U-turn, the authors categorized not prove to be significant to intersection safety, the
the configuration of the cross street right-turn treatment roadway geometry was directly influenced by these
‘‘zone’’ for each one-way frontage road location. Each factors.
configuration included the following two variables:
Right-turn treatment ‘‘zone’’ entrance (see U-Turn Leg Dimensions (Widths and Lengths). Owing to the
Figure 5), and orientation of the U-turns, interchange bridges, and simi-
Right-turn treatment ‘‘zone’’ exit (see Figure 6). lar characteristics, the authors acquired, where feasible,
the turning radius for the U-turn, the turning bay length,
The entrance treatments included shared and exclusive and the lane length. Although the authors evaluated the
cross street right-turn lanes with and without channeliza- significance of all of these candidate variables, the mini-
tion islands. Ultimately, the authors tested the influence mum U-turn radius proved to be the only critical variable
of these entrance treatments and determined that only that influenced safety performance.
Option A (a shared right with no island) was significantly
associated with fewer crashes.
The exit treatment configurations required vehicles to Analysis and Results
merge into existing travel lanes or the geometric config-
The authors performed a statistical analysis for the total
uration included an additional lane (either as an accelera-
crash condition (at turnaround locations) as well as the
tion lane or as another full lane). Even though most of
injury crash condition (denoted as the KAB model in
the study locations had traffic signals, exit treatments at
which ‘‘K’’ represents a fatal crash, ‘‘A’’ represents an
locations with islands often included supplemental traffic
incapacitating injury crash, and ‘‘B’’ represents a non-
control (usually a yield or stop sign).
incapacitating injury crash). The analytical approach
used for this analysis was consistent with that used in a
Depressed or Elevated Interchange Configuration. Of the 336 variety of similar studies (3–5). The following section first
half site locations, 300 (150 of 168 sites) of them had reviews the descriptive statistics for the larger dataset,
6 Transportation Research Record 00(0)
Figure 5. Right-turn treatment ‘‘zone’’ entrance options freeway crashes at the entire diamond interchange (rather
(a) shared right, no island, (b) exclusive right, no island, (c) than crashes on just one side).
exclusive right, painted island, (d) exclusive right, raised island, and
(e) shared right, raised island, large radius.
Descriptive Statistics
Based on the premise that the number of left-turn crashes
then identifies some of the statistical modelling challenges that originated on a frontage road can be expected to be
addressed, and finally presents the resulting statistical reduced at U-turn locations, the authors inspected the
models. Both models predicted crashes for the non- crash data (for the years 2009–2015) and site data using a
Dixon et al 7
variety of target crash conditions including the propor- The authors used a negative binomial model approach
tion of crashes in which the vehicle originated on the for this analysis.
frontage road. In some cases, missing data required slight
modifications in sample sizes. This occurred when a data
element could not be determined using the aerial photos Total Crash Model. The authors used a ‘‘sample expansion’’
and the roadway inventory file. For the initial inspection, step before switching subsets of potential independent vari-
the data included 2019 site-periods (seven potential years ables as the initial model development approach (6–8). For
per site) from 164 sites; however, 77 of the site-periods the development of the final predictive model, the authors
did not have available speed limit data and so the authors focused on signalized intersection locations with speed lim-
could not evaluate these sites relative to speed limit. This its greater than 30 mph. Through the application of step-
resulted in 1016 site-periods from 152 sites with turn- wise regression procedures, the authors then developed a
arounds (see Table 1). final predictive model for total crashes. The model selec-
tion occurred in several stages so that the authors could
consider groups of variables jointly available for data sub-
Statistical Analysis sets. Once a modelling stage had arrived at the point with
Following the initial data inspection, the authors per- the best fit and the fewest variables (referred to as a parsi-
formed a stepwise regression analysis to assess the influ- monious model), the authors fit the model to the largest
ence of significant variables on the total number of subset of data that included all the variables in the model
crashes, as well as crash severity for the dedicated U-turn specification. This iterative process continued until the
lane locations. During this process, the authors noted authors had considered all variables at least two times for
that only a small number of the interchanges had a inclusion into the model. The resulting dataset represented
posted speed limit of 30 mph. Most of the intersections 86 site locations with 459 site-periods available for model
had speed limits of 35 to 55 mph with an average posted estimation. Table 2 shows the final total crash model. The
speed limit of 45 to 46 mph. None of the 30 mph intersec- authors applied scaling adjustments to two of the variables
tions had traffic signals for both of the intersections with more widely dispersed data (9–11). The cross street
associated with the interchange. To mitigate the influ- right-turn entrance treatment A (shared right with no
ence of this subset of study sites, the authors elected to island) proved to be more significant that the other
remove these intersections from the subsequent analysis. entrance treatments. The cross street right-turn exit treat-
ments were not statistically significant for the total crash
model.
Model Development. The crash model development A common goodness of fit assessment gives the cumu-
focused on the identification of statistically significant lative residual (CURE) plot. Optimally, the CURE plot
variables and an assessment of the best functional form for each variable should oscillate around the line that
that represents each identified variable. For example, the represents zero. For the total crash model, these plots
use of logarithmic adjustments may be appropriate for depict minimal deviations beyond the expected bound-
some variables that do not follow a linear format when aries for key variables (see Figure 7). As can be observed,
graphically plotted. The authors used generalized mixed- the expected oscillation occurs for the lower volume
effects models to compensate for more than one study thresholds, but the cumulative residuals appear to be
year of data at a specific site location. The authors also skewed for the higher volume locations.
explored an extensive number of variables during the The final model presents a functional form that incor-
variable selection. Of particular note is the selected traf- porates scaling of some variables. This total crash model
fic volume variable. It is feasible to consider freeway is
13, 516:82
NTotal = e½5:304ð0:271 3 RtAÞ + ð0:703 3 AvgLnÞð0:268 3 Þ + 0:113 3 ½CSAADT
DWY196:319
155:752 10, 059:57 (0:951 3 ln(Rmin))
NTotal =
Rmin0:9512
8 Transportation Research Record 00(0)
Note: Min = minimum; Max = maximum; SD = standard deviation; N = number; RtA = Number of instances at the site where RtA had a shared right turn
lane and no channelization island (see Figure 5). Value of RtA ranges from zero (no shared lane option) to two (shared lane option at both cross street
right-turn locations).
a
Includes adjustment random effects.
b
Significance levels are as follows:
*
Statistically different from 0.0 at the 5.0% significance level.
**
Statistically different from 0.0 at the 1.0% significance level.
***
Statistically different from 0.0 at the 0.1% significance level.
One interesting observation about the final (and ulti- For frontage roads with 2–4 lanes, the number of
mately the KAB) model is that the presence of a turn- crashes increases by a factor of 2.01 (doubles) for
around does not appear as a critical variable in the each additional frontage road lane (calculated as
model. The authors included this variable in the stepwise e(0.7027) = 2.014). This finding is significant at
analysis and it was not significant. This finding suggests 0.1%. This finding is likely a surrogate for the
that constructing a turnaround does not significantly varying frontage road AADT values.
affect the total number of non-freeway interchange The number of crashes reduces as the distance to
crashes, and so this treatment should complement opera- the closest downstream driveway increases. This
tional benefits of adding turnarounds at diamond inter- reduction is approximately 1.7% for each addi-
change locations. tional 10 feet between the closest U-turn exit and
By inspection of the variables included in the total the downstream driveway (calculated as
0:2684
crash model for interchanges with turnarounds, the fol- 1 e 155:752 3 10 = 0:0171). This finding is significant
lowing general observations merit consideration: at 0.1%.
The number of crashes increases by 1.1% for each
Locations where the right-turn from the cross additional 1000 vehicles per day (vpd) increase in
street originates from a shared lane and does not cross street AADT (calculated as
0:1131
have a large turning radius or a raised island 1 e10, 059:6 3 1000 = 0:011). This finding is significant
(Option A) could result in 23.7% fewer crashes at 5.0%.
(calculated as 12e(20.2708) = 0.237). This finding The number of crashes decreases by 8.7% for each
is significant at 1.0%. increase in 10% in the turning radius of the
Dixon et al 9
200
200
Number of Crashes
Number of Crashes
100
100
0
0
-100
-100
-200
-200
0 10000 20000 30000 40000 50000 1.0 1.5 2.0 2.5 3.0
200
Number of Crashes
Number of Crashes
100
100
0
0
-100
-100
-200
-200
D_to_Closest_Driveway MinR
U-Turn (calculated as 1 e0:9512 3 lnð1:1Þ = Sites where the right-turn from the cross street
0:0867). This finding is significant at 0.1%. must share a lane have 26.4% fewer KAB crashes
(calculated as 12e(20.3063)). This finding is signifi-
cant at 1.0%.
KAB Crash Model. The authors used a similar modelling Sites where the right-turn traffic must merge with
approach as that applied for total crashes to evaluate the the frontage road traffic (without adding a lane)
injury crashes at the Texas turnaround locations (see have fewer severe crashes (significant at \
Table 3). The CURE plots for this model exhibited simi- 0.01%). This merge configuration is associated
lar oscillating patterns to those as shown for total crashes with a reduction of 16.4% (calculated as
and therefore are not shown for this model. 12e(20.1801)). This finding is significant at 1.0%.
The KAB crash model equation is:
After simplification, the resulting predictive KAB The number of KAB crashes is smaller by 1.3%
equation takes the following form: for each additional 10 feet between the closest
CS AADT 0:0719
NKAB = 3 e½3:32ð0:31 3 RtAÞ + ð0:48 3 AvgLnÞð0:18 3 MergeRT Þð0:0013 3 DWY Þ
Rmin0:6684
By inspection of the variables included in the KAB downstream driveway and the U-turn exit (calcu-
model for signalized interchanges with turnarounds, the 0:1969
lated as 1 e 151:86 3 10 = 0:0126). This finding is
following observations merit consideration: significant at 1.0%.
10 Transportation Research Record 00(0)
CS_AADT Cross street AADT (vpd) 13,872.4 10,442.3 200 54,309 6,478,395 467
AvgLn Average number of frontage 2.62 0.45 2 4 1225 467
road lanes FRA and FRB (see
Figure 4)
DWY Distance to closest downstream 191.54 151.86 10 500 89,467 467
driveway (ft)
Rmin Minimum turning radius in U- 49.99 14.84 22 129 23,344 467
turn (ft)
Note: Min = minimum; Max = maximum; SD = standard deviation; N = number; RtA = Number of instances at the site where right-turn ‘‘zone’’ entrance
treatment had a shared right turn lane and no channelization island (see Figure 5). Value of RtA ranges from zero (no shared lane option) to two (shared
lane option at both cross street right-turn locations). MergeRT = Number of instances at the site where the right-turn ‘‘zone’’ exit treatment merged into
an existing lane (see Figure 6). Value of MergeRT ranges from zero (only included added lanes) to two (all cross street right-turn lanes require vehicles to
merge into an existing lane).
a
Includes adjustment random effects.
b
Significance levels are as follows:
*
Statistically different from 0.0 at the 5.0% significance level.
**
Statistically different from 0.0 at the 1.0% significance level.
***
Statistically different from 0.0 at the 0.1% significance level.
Significant influence
Site or traffic characteristic Total crashes KAB crashes
As the cross street AADT increases: Total crashes increase KAB crashes increase
Cross street right-turn maneuvers onto the Fewer total crashes Fewer KAB crashes
frontage road that originate in a shared lane
are associate with:
Cross street right-turning vehicles that merge No significantly noticeable change in total crashes Fewer KAB crashes
into existing lanes are associated with:
As the number of lanes increases for each Total crashes increase KAB crashes increase
frontage road approach:
As the longitudinal distance between the U-turn Total crashes decrease KAB crashes decrease
exit to the nearest downstream driveway
increases:
As the minimum U-turn radius increases: Total crashes decrease KAB crashes decrease